Abstract 4348795: Elevated LDL-C and Fasting Plasma Glucose Predict Benefit from Invasive Strategy in Chronic Coronary Disease: A Causal Forest Analysis of the ISCHEMIA Trial

Abstract

The ISCHEMIA trial showed no significant difference in the rate of primary outcome events between patients with chronic coronary artery disease who used invasive strategies (INV) versus conservative strategies (CON). In this study, we used causal forest analysis to identify potentially heterogeneous treatment effects using half of the trial data.We constructed 1000 trees and the minimum number of observations of leaf nodes in each tree is 60. The remaining data were used as an independent test set using Cox proportional hazards regression modeling to test for potentially heterogeneous treatment effects. Of 5179 participants in ISCHEMIA, 1220 participants (23.5%) were LDL-C > 86mg/dl and fasting plasma glucose > 104mg/dl, with 595 of 1220 participants (48.8%) randomized to the training data set and 624 of 1220 participants (51.2%) randomized to the testing data set; 914 of 1220 participants (74.9%) were male, with the mean age of 63.8 ±9.5 years and 187 of 1220 (15.3%) participants experienced a primary outcome event. Using causal forest model, we identified subgroups with possible heterogeneity in INV. In the testing data, Cox models revealed that a subgroup (LDL-C＞86 mg/dl with FPG＞104 mg/dl) had a significantly higher frequency of primary outcome events in CON (18.1% [56 of 309]) than in INV (12.1% [38 of 315]), with a HR of 0.66 (95% CI, 0.44-0.99; P=0.045). Finally, we found that chronic coronary disease patients with LDL-C＞86 mg/dl and FPG＞104 mg/dl can achieve more benefit from INV, the finding that provides a new direction for individualized therapy.